Contrary to so-called garden path sentences (Ferreira, Christianson, & Hollingworth, 2001; Osterhout & Holcomb, 1992), the violation and the re-analysis of GP jokes are localized at the semantic rather than syntactic level. Here, the mental representation of the discourse, theoretically depicted as mental model (Johnson-Laird, 1983) or situation model (Kintsch, 1988, 1998), is violated at the PL. It is commonly assumed that the discourse comprehension is an active process of cognitive construction that involves the integration of explicit linguistic input with other linguistic and non-linguistic context information, including new semantic and pragmatic inferences and knowledge from long-term memory. Most importantly, a committed false belief concerning the interpretation of the text has to be substituted. This “(belief) revision” of the semantic representation (of SU) is the crucial mechanism during the comprehension of GP jokes (cf. Mayerhofer
& Schacht, 2013). Consider again example (1).
(1) “Mummy, I just turned 14 years. May I please, finally, be allowed to wear a bra and make up?” -“No, and eat up your soup, my son.”
Given the linguistic information and the recipient’s world knowledge, the child being a girl is the most plausible interpretation of the set-up phase. This interpretation gets violated when one hears the mother calling the child “son” (PL), thus leading to
incongruity. Belief revision occurs, and the recipient represents a boy who would love to wear a bra and make up. This incongruity resolution, in combination with the activation of the alternative, hidden interpretation and with its “inappropriateness” (Ritchie, 2004, p.61), is typically accompanied by the experience of laughter and mirth in the recipient.
Many researchers agree upon the outlined sequential comprehension process, supported by empirical evidence. Vaid et al. (2003) demonstrated priming e↵ects due to the dom-inant semantic networks specifically activated at di↵erent stages of joke comprehension over time. Coulson and Kutas (1998) found longer reading times for joke endings com-pared to straight (coherent) endings. These longer reading times were also accompanied by regressive eye movements after reading of the punch-line (Coulson et al., 2006).
Evidence for the enhanced costs of semantic revision also comes from non-joke texts (Carreiras, 1996; Sturt, 2007).
Recently, several studies using event-related brain potentials (ERPs) have investigated the processing of jokes and verbal humor. Three (groups of) ERP components were especially fruitful for the study of verbal humor: the N400, late positivities, and the left anterior negativity. The N400 component (Kutas & Hillyard, 1980) is an enhanced negative-going deflection at cento-parietal electrodes starting around 200–250 ms after stimulus onset and lasting until 500–550 ms after stimulus onset with a peak around 400 ms, hence the name. It reliably occurs with semantic violations during sentence or discourse comprehension (Berkum, Hagoort, & Brown, 1999). Other important fac-tors that influence the amplitude of the N400 component are the predictability of a word in a given context, as for example reflected by the Cloze-probability (Kutas &
Hillyard, 1984), and the semantic relatedness between the context and the expected word. The N400 e↵ect functionally reflects semantic integration difficulties at the inter-face of word/stimulus recognition, linguistic and nonlinguistic context, and conceptual binding with the long-term-memory during an active comprehension process (Kutas &
Federmeier, 2011). Previous ERP studies on joke comprehension have led to hetero-geneous evidence regarding N400 e↵ects. Derks, Gillikin, Bartolome-Rull, and Bogart (1997) found augmented N400 amplitudes for jokes that also elicited a higher activation of the zygomatic muscle, indicating the elicitation of positive emotions. Coulson and Kutas (2001) found an N400 e↵ect for joke endings involving frame shifting compared to straight endings. The e↵ect was restricted to jokes with high semantic constraint on the ending. This finding was replicated in follow-up studies, shown to be only present for participants with a low verbal intelligence score (Coulson & Lovett, 2004), and to be related to the visual field of the stimulus presentation (Coulson & Williams, 2005).
Several ERP studies on language comprehension demonstrated syntactic violations to elicit an augmented positivity at posterior scalp sites. This so-called P600 component usually starts around 600 ms after stimulus onset and lasts until around 1200 ms. Since these late positivities are especially triggered by syntactic anomalies, such as in GP sentences (Bever, 1970; Osterhout & Holcomb, 1992), they are commonly considered to reflect syntactic repair processes which occur after the detection of a syntactic violation for the initially dominant parsing. However, Van Herten et al. (2005) found posterior P600 e↵ects for semantic anomalies, and experimentally ruled out the possibility of a hidden syntactic anomaly being responsible for the component. This finding led the authors to argue that the P600 is a form of monitoring component “that checks upon the veridicality of one’s analysis” (Van Herten et al., 2005, p.254). In line with this assumption, the P600 has been suggested to reflect a combinatorial process, integrating both syntactic and semantic features of a sentence (e.g., Wicha, Moreno, & Kutas, 2004; Mart´ın-Loeches et al., 2009), and has also been reported for increased discourse complexity (Burkhardt, 2007). Moreover, a late positivity e↵ect – distinguishable from the typical P600 e↵ect by its frontal distribution – has been reported (Schacht et al., 2010) and related to the complexity and the ambiguity of a text (Kaan & Swaab, 2003).
In many regards, GP jokes might be assumed as a semantic equivalent of GP sentences.
Thus, the question is whether a semantic repair process in jokes – such as the belief re-vision – triggers similar brain response patterns as the mainly syntactic repair processes (P600 at posterior sites). Previous evidence has partly indicated such similarity, but remains incomplete (Coulson & Lovett, 2004; Marinkovic et al., 2011).
Apart from the P600 like findings, there is strong evidence that joke endings, triggering believe-revision processes, elicit a left-lateralized sustained anterior negativity (Late Left Anterior Negativity; LLAN), between 500 and 900 ms after stimulus onset. This component has been shown only for good comprehenders (Coulson & Kutas, 2001;
Coulson & Williams, 2005) or restricted to left-handed participants (Coulson & Lovett, 2004). Coulson and co-workers suggested that the component reflects the successful comprehension of jokes and called this e↵ect “frame-shifting component” according to their terminology. The LLAN has also been considered to reflect working memory activity necessary for the computation of a new mental representation of the discourse (Baggio, Van Lambalgen, & Hagoort, 2008; Meltzer & Braun, 2013; M¨unte, Schiltz, &
Kutas, 1998).
GP jokes also reliably lead to the subjective experience of mirth. Therefore, one might expect other ERP components elicited by jokes, reflecting the emotional processes.
Emotion-related ERP responses to humorous visual stimuli have been reported as Pos-terior Positivities between 300 and 600 ms after the onset (Gierych, Milner, & Michalski, 2005; Korb, Grandjean, Samson, Delplanque, & Scherer, 2012). These components show strong similarities to the late positive complex (LPC), which has repeatedly been shown in response to emotional stimuli, such as a↵ective pictures (e.g., Cuthbert, Schupp, Bradley, Birbaumer, & Lang, 2000; Schupp et al., 2000), and to facial emotional ex-pressions and emotional words (e.g., Schacht & Sommer, 2009a, 2009b). This e↵ect has been related to sustained, elaborative processing of emotional relevance of a given
stim-ulus. At longer latencies, Du et al. (2013) reported an enhanced positivity to Chinese jokes compared with neutral Chinese texts between 1250 and 1400 ms after the stimulus onset, which the authors related to an a↵ective stage of the joke processing.
It is the main aim of the present study to disentangle di↵erent sub-processes or process-ing stages, respectively, involved in the comprehension of GP jokes to be reflected in distinguishable ERP components over time. At least, three di↵erent processing stages are hypothesized to be involved: (a) the violation of the pre-dominant initial semantic representation, (b) the revision of this semantic representation, and (c) the occurrence of an emotional reaction. To this aim, we constructed parallel versions of selected jokes in such a way that all comprehension processes should remain constant apart from the processes of interest outlined above. This manipulation was realized by exchanging only the final word of the original jokes as in the following examples (compared to (1)):
(2) “Mummy, I just turned 14 years. May I please, finally, be allowed to wear a bra and make up?” -“No, and eat up your soup, my girl.”
(3) “Mummy, I just turned 14 years. May I please, finally, be allowed to wear a bra and make up?” -“No, and eat up your soup, my father.”
In example (2), the interpretation of the whole discourse is straight-forward and co-herent. Thus, no belief revision is necessary. In example (3), the initial interpretation gets violated. The final sentence is a grammatically and semantically correct sentence, but its final word is discourse incoherent, thus triggering revision processes. In contrast to the joke ending of example (1), no hidden interpretation (or at least no plausible one) can be activated and no alternative meaningful coherent representation of the text can be constructed. This makes the whole text incomprehensible. The joke endings share the discourse incoherence with (3) at the occurrence of the final word (PL), but it shares the comprehensibility of a meaningful discourse with (2), once the belief
re-vision has been successfully carried out. In a series of experiments, we investigated the neuro-cognitive processes being specific for GP jokes, using 48 GP jokes and their coherent and incoherent manipulations as stimuli. Experiment 1 focused on behavioral measures using a self-paced reading time paradigm. Here, I expected increasing read-ing times from coherent over incoherent to joke endread-ings. In Experiment 2 to 4, ERPs were of main interest in order to localize the GP-specific sub-processes. Hypotheses were as follows: Joke endings and incoherent endings both represent the violation of the initially dominant semantic representation and should thus elicit an augmented N400 component. Successful belief-revision processes in GP joke comprehension – requiring enhanced inferential and working-memory related processes – should be reflected in the occurrence of LLAN components. Only joke endings should elicit an emotional response.
Therefore, we expected emotion-related ERP components at subsequent, late stages of joke processing, namely following the violation and the revision processes.
Another potentially fruitful indicator of both cognitive and emotional processes during the comprehension of jokes could be provided by pupillary responses, which we also measured in Experiment 2. First, changes of pupil size have been shown to be a sensi-tive measure for the cognisensi-tive load during a task: Higher cognisensi-tive load leads to larger pupil dilation (Kahneman & Beatty, 1966; Van Der Meer et al., 2010). Second, larger pupil dilations have also been reported in association with higher emotional involve-ment, related to the arousal (Bradley, Miccoli, Escrig, & Lang, 2008) or to the intensity (Partala & Surakka, 2003) of an emotional reaction. Both factors cognitive load and emotional processing have been shown to a↵ect pupil dilations also in the processing of verbal stimuli, such as single word processing and recognition (Bayer et al., 2011;
V˜o et al., 2008). Since the successful comprehension of GP jokes is hypothesized to in-volve both increased cognitive processing e↵ort and an emotional response, we expected larger pupil dilations after joke endings compared to coherent endings. Changes of pupil
size to incoherent endings should be intermediate due to enhanced cognitive demands (violation detection) on the one hand but the absence of both revision processes and emotional response on the other hand.